NumPy append vs Python append

OP intended to start with empty array. So, here’s one approach using NumPy

In [2]: a = np.empty((0,3), int)

In [3]: a
Out[3]: array([], shape=(0L, 3L), dtype=int32)

In [4]: a = np.append(a, [[1,2,3]], axis=0)

In [5]: a
Out[5]: array([[1, 2, 3]])

In [6]: a = np.append(a, [[1,2,3]], axis=0)

In [7]: a
Out[7]:
array([[1, 2, 3],
       [1, 2, 3]])

BUT, if you’re appending in a large number of loops. It’s faster to append list first and convert to array than appending NumPy arrays.

In [8]: %%timeit
   ...: list_a = []
   ...: for _ in xrange(10000):
   ...:     list_a.append([1, 2, 3])
   ...: list_a = np.asarray(list_a)
   ...:
100 loops, best of 3: 5.95 ms per loop

In [9]: %%timeit
   ....: arr_a = np.empty((0, 3), int)
   ....: for _ in xrange(10000):
   ....:     arr_a = np.append(arr_a, np.array([[1,2,3]]), 0)
   ....:
10 loops, best of 3: 110 ms per loop